case wrote:
> The recent change of the (fairly simple) resp program is creating problems.
> First, the resp_modules.o file created by gfortran on MacOSX 10.5 is over
> 2 Gbytes in length. And then my laptop can't malloc enough memory to
> make the resp executable (which also appears to be more than 2 Gbytes.)
>
> Does anyone else see this? I'm tempted to undo the f90 conversion, because
> a benefit of the older code is that it identical to that used by the R.E.D.
> folks, so we won't have to try to debug differences between the two.
>
> ...dac
Hmmmm...
There is a stunning difference between the output produced by the Intel
compiler and that produced by the GNU compiler:
Intel compiler
==============
[roberts.qtpclnt070 etc]$ ls -lb resp_modules.o
-rw-r--r-- 1 roberts admin 5010 10 Feb 18:29 resp_modules.o
GNU compiler
============
[roberts.qtpclnt070 etc]$ ls -lb resp_modules.o
-rw-r--r-- 1 roberts admin 2074740852 10 Feb 18:33 resp_modules.o
Compare with another "modules" object file in the same directory,
nucgen_modules.o:
Intel compiler
==============
[roberts.qtpclnt070 etc]$ ls -lb nucgen_modules.o
-rw-r--r-- 1 roberts admin 1125 10 Feb 18:37 nucgen_modules.o
GNU compiler
============
[roberts.qtpclnt070 etc]$ ls -lb nucgen_modules.o
-rw-r--r-- 1 roberts admin 1144 10 Feb 18:36 nucgen_modules.o
I guess I didn't realise the problem because I have a beast of Mac that
has 10 GB of RAM (so all I noticed was a slow compilation). But still, I
agree that a 2 GB resp executable isn't acceptable, and don't understand
why it should be that way. While I (of course) don't like to see my work
vanish into the ether, by all means roll back if it's the best thing to
do, and I apologise for any inconvenience.
Ben
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Feb 10 2010 - 16:00:02 PST
